1. Field of the Invention
The present invention relates in general to the field of data processing circuitry and, more specifically, to systems and methods for improving data processing circuit performance by providing improved dynamic monitoring management of thermal transients in integrated circuits.
2. Description of the Related Art
Modern integrated circuits have benefited from several decades of Moore's law. Many modern integrated circuits have more than a billion transistors. There are also has been a trend toward highly non-homogeneous, rapidly varying power densities in semiconductor devices, particularly in microprocessors. This has driven the need for improved methods to predict transient temperature responses of systems and devices with multiple heat sources.
While computational resources required for direct thermal simulation, e.g., using detailed finite element modeling (FEM), have historically been used, several methods implementing reduced-complexity models have recently been developed for detection of dynamic thermal responses electronic systems to provide a significant reduction in computing time. However, these approaches still consume considerable computational resources, thus inhibiting implementation of these techniques in real-time temperature prediction and dynamic power management applications.
In recent years, there has been increased interest in the application of dynamic thermal management (DTM), for example, through power regulation when operating temperatures exceed safety thresholds. In many integrated circuits, actual thermal sensors are located at predetermined locations on the integrated circuit. An alternative technique for thermal characterization is to measure a “thermal step-response function (often called the transient thermal impedance). Existing methods for implementing this technique, typically calculate the transient temperature of the systems, avoiding direct finite element numerical simulations. These techniques typically involve the use of thermal equivalent circuits, time-domain step response curves (transfer functions) and discrete convolution integrals.
In existing techniques for implementing thermal step-function measurement, calculation of the time-domain discrete convolution integral requires extensive numerical computational power, even when using discrete Fourier transforms or approximated interpolation of system transfer functions. This is due to the fact that the convolution operation yields the amount of overlap between power excitation and the system thermal transfer function, which is determined not only by values at the current time step, but also by the previous time steps. Therefore, these techniques require the storage of previous data points for current time thermal calculation, where the links of data depend on the characteristics of both the power profile and the thermal transfer function of the system. Thermal time constants of microcircuit components usually span over several orders of magnitude, e.g., from microseconds to minutes. Prior techniques for measuring the thermal response of circuit components having longer time constants make if difficult to respond to thermal changes in real time.
In view of the foregoing, it is apparent that there is a need for improved systems and methods for dynamic monitoring and management of thermal transients in data processing circuits. In particular, it would be desirable to provide a system and method for measuring thermal response of circuit components in a very short period of time.
Where considered appropriate, reference numerals have been repeated among the drawings to represent corresponding or analogous elements.